Double-Talk Robust Acoustic Echo Cancellation Using Partition Block Frequency-Domain Adaptive Filtering

This work aims at introducing a new acoustic echo cancellation algorithm robust to double-talk situation. We propose a method which combines partition-block frequency domain adaptive filtering (PBFDAF) and best linear unbiased estimation (BLUE) through local signal characteristics recursive estimation. We report that our method outperforms an Acoustic Echo Cancellation (AEC) baseline using BLUE for Echo Return Loss Enhancement (ERLE) and show its usefulness when used in an automated speech recognition pipeline. Improvements are obtained for situations featuring either continuous or discontinuous echo signals.